“I’m not convinced Musk has formulated the correct problem statement.”
I’d be shocked if he has.
But he is *trying* to come up with it. Even the method he is using may not be correct… but at least he’s working on it.
Similar with the “going to Mars” thing. We need someone dragging us along, and our politicians need someone *shaming* them into participating (not that they experience it as shame, but if some *other* politicians embraces it, it could hurt their numbers! THE HORROR!).
Loading...
I don’t agree with Elon on this.
The question he posed is a good one, and one that physicists have wanted to answer for centuries. But it’s not the only question worth answering.
As I see it, AI is a tool. At the moment, a not very reliable one, but with potential. And it’s a very flexible tool, potentially capable of doing many things. It’s perfectly sensible for different organizations to focus on different potential applications and how to improve the handling of those.
The history of technology is full of things like this. Consider the steam engine: originally a device invented for pumping water out of mines, it morphed into a power source for ships, and then brought about the creation of railroads, and early cars. Or the computer: originally a machine for solving military problems (like ballistics), then processing census data and bookkeeping, then many more things from cryptography to weather forecasting to engineering design to entertainment. A single organization focused single-mindedly on one of those applications would have entirely missed that broad potential.
Loading...
No one ever accused Elon of playing small ball. Whether that is the right goal or whether it can be achieved are different questions. “the universe is not only stranger than we imagine, it is stranger than we can imagine.” It may be that AI can do the imagining. Or not.
Loading...
“There is a theory which states that if ever anyone discovers exactly what the Universe is for and why it is here, it will instantly disappear and be replaced by something even more bizarre and inexplicable.
There is another theory which states that this has already happened.”
— Douglas Adams, The Restaurant at the End of the Universe
Loading...
I remember that line. 🙂
Loading...
“There is a third theory that both of the previous theories were concocted by a wily editor of the Hitchhiker’s Guide To The Galaxy, in order to increase the general level of fear and anxiety in the universe, and thus boost sales of the Guide.
This last theory is the most convincing.”
HHGTTG, Secondary Phase.
(I have all five of the HHGTTG radio shows in a USB drive plugged into my car. If I don’t have the above quote word-perfect, I’m really close.)
Loading...
Curiously, there was a science-fiction short story by I think Isaac Asimov in which someone decides that jokes are put into our brains as part of a big experiment by aliens to examine various aspects of our minds. This was noted in the story by the observation that everyone “hears a joke” from someone else and no one seems to have made up any jokes except puns, which are funny only to the teller of the pun. The person who figured this out said that once the secret was out, the aliens would stop giving us jokes and we wouldn’t laugh at them any more. This happens in the story and sure enough, no one laughs at jokes, even old ones we already knew.
Loading...
I’m not convinced Musk has formulated the correct problem statement. But I am certain I do not have a better one.
I’d be shocked if he has formulated the “correct” problem statement (for some definition of “correct,” which varies greatly on what problem he’s trying to solve).
But his is worlds better than what seems to be the current standard in the AI “corporate arms race,” which from the outside looks an awful lot like, “What can we do to make the most powerful AI quickest and be first to market with a ‘solution’ nobody fully knows how to use, let alone control?”
And as a secondary follow-up, “What can we do to use the tool we have to make more money in the background?”
From what I can see, none of the companies developing powerful AI models — not a single one — really know what they’re going to do with it. Oh, they have a few ideas, but none of them really require even a fraction of the capability they’re building. The only clear motivation I see is the one to NOT be late to market.
It’s like a guy who says he wants to build the biggest, most powerful monster truck ever, with the biggest engine, the biggest tires, the heftiest frame and suspension to support it all, etc. When his wife asks, “Okay, but what will you DO with that?” His answer is, “I dunno. Drive around? Photo shoots? TikTok videos?”
Building all that capability just for the purpose of building the capability, and no clear idea what it will be used (or useful) for. That’s what it looks like to me, from outside. At least Elon has a big idea that doesn’t resemble an Underpants Gnomes scheme.
Loading...
The distinction between solving a benchmark and formulating the right underlying problem is important. In AI development, model capability matters, but so do the questions, constraints, and evaluation criteria chosen before the system is built.
“I’m not convinced Musk has formulated the correct problem statement.”
I’d be shocked if he has.
But he is *trying* to come up with it. Even the method he is using may not be correct… but at least he’s working on it.
Similar with the “going to Mars” thing. We need someone dragging us along, and our politicians need someone *shaming* them into participating (not that they experience it as shame, but if some *other* politicians embraces it, it could hurt their numbers! THE HORROR!).
I don’t agree with Elon on this.
The question he posed is a good one, and one that physicists have wanted to answer for centuries. But it’s not the only question worth answering.
As I see it, AI is a tool. At the moment, a not very reliable one, but with potential. And it’s a very flexible tool, potentially capable of doing many things. It’s perfectly sensible for different organizations to focus on different potential applications and how to improve the handling of those.
The history of technology is full of things like this. Consider the steam engine: originally a device invented for pumping water out of mines, it morphed into a power source for ships, and then brought about the creation of railroads, and early cars. Or the computer: originally a machine for solving military problems (like ballistics), then processing census data and bookkeeping, then many more things from cryptography to weather forecasting to engineering design to entertainment. A single organization focused single-mindedly on one of those applications would have entirely missed that broad potential.
No one ever accused Elon of playing small ball. Whether that is the right goal or whether it can be achieved are different questions. “the universe is not only stranger than we imagine, it is stranger than we can imagine.” It may be that AI can do the imagining. Or not.
“There is a theory which states that if ever anyone discovers exactly what the Universe is for and why it is here, it will instantly disappear and be replaced by something even more bizarre and inexplicable.
There is another theory which states that this has already happened.”
— Douglas Adams, The Restaurant at the End of the Universe
I remember that line. 🙂
“There is a third theory that both of the previous theories were concocted by a wily editor of the Hitchhiker’s Guide To The Galaxy, in order to increase the general level of fear and anxiety in the universe, and thus boost sales of the Guide.
This last theory is the most convincing.”
HHGTTG, Secondary Phase.
(I have all five of the HHGTTG radio shows in a USB drive plugged into my car. If I don’t have the above quote word-perfect, I’m really close.)
Curiously, there was a science-fiction short story by I think Isaac Asimov in which someone decides that jokes are put into our brains as part of a big experiment by aliens to examine various aspects of our minds. This was noted in the story by the observation that everyone “hears a joke” from someone else and no one seems to have made up any jokes except puns, which are funny only to the teller of the pun. The person who figured this out said that once the secret was out, the aliens would stop giving us jokes and we wouldn’t laugh at them any more. This happens in the story and sure enough, no one laughs at jokes, even old ones we already knew.
I’m not convinced Musk has formulated the correct problem statement. But I am certain I do not have a better one.
I’d be shocked if he has formulated the “correct” problem statement (for some definition of “correct,” which varies greatly on what problem he’s trying to solve).
But his is worlds better than what seems to be the current standard in the AI “corporate arms race,” which from the outside looks an awful lot like, “What can we do to make the most powerful AI quickest and be first to market with a ‘solution’ nobody fully knows how to use, let alone control?”
And as a secondary follow-up, “What can we do to use the tool we have to make more money in the background?”
From what I can see, none of the companies developing powerful AI models — not a single one — really know what they’re going to do with it. Oh, they have a few ideas, but none of them really require even a fraction of the capability they’re building. The only clear motivation I see is the one to NOT be late to market.
It’s like a guy who says he wants to build the biggest, most powerful monster truck ever, with the biggest engine, the biggest tires, the heftiest frame and suspension to support it all, etc. When his wife asks, “Okay, but what will you DO with that?” His answer is, “I dunno. Drive around? Photo shoots? TikTok videos?”
Building all that capability just for the purpose of building the capability, and no clear idea what it will be used (or useful) for. That’s what it looks like to me, from outside. At least Elon has a big idea that doesn’t resemble an Underpants Gnomes scheme.
The distinction between solving a benchmark and formulating the right underlying problem is important. In AI development, model capability matters, but so do the questions, constraints, and evaluation criteria chosen before the system is built.